Skip to content

LLM inference in C/C++

C++ ggml

gguf : add tensor shape accessor (#24405)

QuintinShawad8d8219915d
diff --git a/ggml/include/gguf.h b/ggml/include/gguf.h
index 67851ba6f1..b3a1e1230a 100644
--- a/ggml/include/gguf.h
+++ b/ggml/include/gguf.h
@@ -125,12 +125,13 @@ extern "C" {
// get ith C string from array with given key_id
GGML_API const char * gguf_get_arr_str (const struct gguf_context * ctx, int64_t key_id, size_t i);
- GGML_API int64_t gguf_get_n_tensors (const struct gguf_context * ctx);
- GGML_API int64_t gguf_find_tensor (const struct gguf_context * ctx, const char * name); // returns -1 if the tensor is not found
- GGML_API size_t gguf_get_tensor_offset(const struct gguf_context * ctx, int64_t tensor_id);
- GGML_API const char * gguf_get_tensor_name (const struct gguf_context * ctx, int64_t tensor_id);
- GGML_API enum ggml_type gguf_get_tensor_type (const struct gguf_context * ctx, int64_t tensor_id);
- GGML_API size_t gguf_get_tensor_size (const struct gguf_context * ctx, int64_t tensor_id);
+ GGML_API int64_t gguf_get_n_tensors (const struct gguf_context * ctx);
+ GGML_API int64_t gguf_find_tensor (const struct gguf_context * ctx, const char * name); // returns -1 if the tensor is not found
+ GGML_API size_t gguf_get_tensor_offset(const struct gguf_context * ctx, int64_t tensor_id);
+ GGML_API const char * gguf_get_tensor_name (const struct gguf_context * ctx, int64_t tensor_id);
+ GGML_API const int64_t * gguf_get_tensor_ne (const struct gguf_context * ctx, int64_t tensor_id); // returns ne, an array of GGML_MAX_DIMS elements; ne[dim] is 1 for dim >= n_dims
+ GGML_API enum ggml_type gguf_get_tensor_type (const struct gguf_context * ctx, int64_t tensor_id);
+ GGML_API size_t gguf_get_tensor_size (const struct gguf_context * ctx, int64_t tensor_id);
// removes key if it exists, returns id that the key had prior to removal (-1 if it didn't exist)
GGML_API int64_t gguf_remove_key(struct gguf_context * ctx, const char * key);
diff --git a/ggml/src/gguf.cpp b/ggml/src/gguf.cpp
index c3ffa1a134..7920b8100b 100644
--- a/ggml/src/gguf.cpp
+++ b/ggml/src/gguf.cpp
@@ -1186,6 +1186,11 @@ const char * gguf_get_tensor_name(const struct gguf_context * ctx, int64_t tenso
return ctx->info[tensor_id].t.name;
}
+const int64_t * gguf_get_tensor_ne(const struct gguf_context * ctx, int64_t tensor_id) {
+ GGML_ASSERT(tensor_id >= 0 && tensor_id < gguf_get_n_tensors(ctx));
+ return ctx->info[tensor_id].t.ne;
+}
+
enum ggml_type gguf_get_tensor_type(const struct gguf_context * ctx, int64_t tensor_id) {
GGML_ASSERT(tensor_id >= 0 && tensor_id < gguf_get_n_tensors(ctx));
return ctx->info[tensor_id].t.type;
diff --git a/tests/test-gguf.cpp b/tests/test-gguf.cpp
index ddb1b4d948..2875dec806 100644
--- a/tests/test-gguf.cpp
+++ b/tests/test-gguf.cpp
@@ -662,6 +662,13 @@ static bool handcrafted_check_tensors(const gguf_context * gguf_ctx, const unsig
if (gguf_get_tensor_type(gguf_ctx, id) != type) {
ok = false;
}
+
+ const int64_t * ne = gguf_get_tensor_ne(gguf_ctx, id);
+ for (int j = 0; j < GGML_MAX_DIMS; ++j) {
+ if (ne[j] != shape[j]) {
+ ok = false;
+ }
+ }
} else {
ok = false;
continue;